No Raid Report, Just the Machinery

There is a small but decisive problem with an ICE-in-Minnesota newsletter built from this packet: it contains no reporting, court filing, enforcement data, agency statement, or local account about ICE, Minnesota, or a relevant federal action. That is not a clerical nuisance. It is the line between journalism and the sort of content mill that can manufacture certainty from a headline, three vibes, and an aggressively caffeinated autocomplete bar.

The materials do contain two stories about systems that turn ordinary infrastructure into machinery for observation, manipulation, or profit: a security investigation into compromised streaming boxes and OpenAI’s account of making advanced models cheaper and faster. Neither establishes anything about immigration enforcement. The responsible conclusion is therefore narrow: the evidentiary record supplied here cannot support claims about what ICE is doing in Minnesota, who is affected, or what legal authority is being used.

That restraint matters especially in civil-liberties reporting. An absence of evidence is not proof of innocence, guilt, competence, or conspiracy; it is simply an absence of evidence. Before declaring a raid, a pattern, or a constitutional crisis, one needs the unglamorous furniture of accountable reporting: dates, locations, records, named institutions, and sources who can be checked. The republic remains frustratingly dependent on paperwork.

What the packet does show is how readily a mundane household object can become somebody else’s infrastructure. KrebsOnSecurity’s report describes Bitsight’s analysis of an expired domain used by H96 streaming devices. The analysis found tens of thousands of boxes reporting hardware details and installed-app lists while presenting themselves as mobile phones.

Bitsight attributed the operation to applications associated with Zhejiang Fengwo IoT Technology and described devices alternating between residential-proxy activity and automated advertising fraud. It tracked roughly 38,000 boxes contacting one older domain and estimated nearly $50,000 a day in ad-fraud revenue from that portion of the operation, excluding any proxy income. Those are findings and estimates attributed to Bitsight, not an independently reproduced forensic record; the distinction is important, even when the alleged scheme has the unmistakable aroma of a digital roach motel.

The civil-liberties lesson is not that every cheap streaming stick is a miniature surveillance state. The report explicitly does not establish that every H96 unit, or every generic Android device sold by major retailers, is implicated. It is that consumers routinely inherit opaque software supply chains—and their network, device, and behavioral data can become a commercial asset without a meaningful moment of informed consent. The practical danger begins long before anyone puts on a uniform: it begins when the thing in your home reports to someone you have never heard of.

Meanwhile, OpenAI says it has cut GPT-5.6 Luna’s API price by 80% and Terra’s by 20%, while adding a faster, higher-priced processing option for Sol. Luna is priced at $0.20 per million input tokens and $1.20 per million output tokens; Terra at $2 and $12. The company presents these changes as the result of model, inference, context-management, and production-engineering improvements.

Its more consequential claim is not merely that a chatbot got cheaper. OpenAI says its own model helped optimize production kernels, run hundreds of token-generation experiments, and reduce end-to-end serving cost by 20%. Such claims may be true, useful, or both. But they come from the vendor’s product announcement, alongside internal and customer evaluation claims whose full methods are not supplied. A price list is admirably legible; a model’s actual institutional impact is rarely so considerate.

For public power—including any future investigation into immigration enforcement—the relevant question is not whether AI is clever enough to summarize a case file. It is whether an institution can show what data entered the system, who authorized its use, what the model produced, how errors were caught, and whether a person can challenge the result. Cheaper automated reasoning lowers the cost of doing things at scale. It does not lower the constitutional price of getting them wrong. When intelligence becomes inexpensive, verification becomes the luxury good we cannot afford to treat as optional.